Power Amplifier Self-Optimization via Neural Network Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current power amplifier technologies face challenges in achieving high efficiency and linearity, especially in Massive MIMO systems operating at millimeter wave frequencies, as they are prone to energy losses and sensitive to physical construction, and existing optimization methods are either costly or rely on basic measurements that do not provide comprehensive performance information.
Innovation Solution
A wireless communication system with a power amplifier and a sensor subsystem that performs asynchronous statistical sampling of RF input and output signals, using a neural network processor to infer performance metrics and control internal parameters for optimization, allowing for scalable, low-power, and cost-effective linearity and efficiency improvements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If Envelope Tracking (ET) or Digital Predistortion (DPD) is used to improve power amplifier efficiency, then efficiency is improved, but the approach does not scale to meet Massive MIMO and higher bandwidth requirements
Solution Approach 1:
The power amplifier system performs self-diagnosis and self-optimization by using its own output signal to generate performance measurements through the sensor subsystem and neural network processor, eliminating the need for external measurement equipment and enabling autonomous adaptation to different operating conditions and bandwidth requirements
Solution Approach 2:
The system dynamically adjusts internal power amplifier parameters based on real-time performance measurements and neural network inference, allowing the amplifier to adapt its operating characteristics to meet varying bandwidth and efficiency requirements across different Massive MIMO configurations
2Measurement precision
If external equipment is used to generate performance measurements, then measurement accuracy is improved, but the cost becomes prohibitively expensive
Solution Approach 1:
The system creates a virtual copy of the measurement function by using the power amplifier's own output signal and a sensor subsystem to generate performance measurements that replicate what expensive external equipment would provide, but at a fraction of the cost through software-based neural network processing
Solution Approach 2:
The sensor subsystem acts as an intermediary between the power amplifier output and the neural network processor, converting the RF output signal into measurable performance metrics that the neural network can process to infer amplifier characteristics without requiring expensive external measurement devices
3Device complexity
If basic measurements such as temperature, DC currents and average power are used, then device complexity is reduced, but comprehensive performance information is not provided
Solution Approach 1:
The system implements a feedback loop where the sensor subsystem continuously monitors the power amplifier output, the neural network processor analyzes the measurements to infer performance metrics, and the results are used to optimize amplifier operation, providing comprehensive performance information while maintaining relatively simple hardware
Data Source
AI summary
A wireless communication system includes a power amplifier (PA) configured to receive a radio frequency (RF) input signal and to produce a PA output signal, the PA output signal being an amplified version of the RF input signal. A sensor subsystem is configured to perform asynchronous statistical sampling of the RF input signal and of the PA output signal and to generate a sensor subsystem output. A controller, in communication with the sensor subsystem, is configured to obtain the sensor subsystem output and to infer performance of the PA, and may control one or more of a plurality of internal PA parameters. The controller may include a neural network processor to associate a particular statistical input/output characterization with a particular inferred performance for the PA. Compared to known approaches, the system is scalable and achieves lower power consumption, and is configured to obtain information about linearity performance.